Energy Policy 51 (2012) 642–651

Contents lists available at SciVerse ScienceDirect

Energy Policy
journal homepage: www.elsevier.com/locate/enpol

Storage and balancing synergies in a fully or highly renewable

pan-European power system
Morten Grud Rasmussen a,b,c,n, Gorm Bruun Andresen b,c, Martin Greiner b,c
a
Center for Mathematical Sciences, Technische Universität München, Boltzmannstraße 3, 85748 Garching, Germany
b
Department of Engineering, Aarhus University, Ny Munkegade 118, 8000 Aarhus C, Denmark
c
Department of Mathematics, Aarhus University, Ny Munkegade 118, 8000 Aarhus C, Denmark

H I G H L I G H T S

c We model a wind and solar based European power system with storage and balancing.
c We find that storage needs peaks when average renewable generation matches load.
c We find strong synergetic effects when combining storage and balancing.
c We study the effects of a storage capable of storing 6 h average use.
c We find a realisable fully renewable scenario based on wind, solar and hydro power.

a r t i c l e i n f o a b s t r a c t

Article history: Through a parametric time-series analysis of 8 years of hourly data, we quantify the storage size and
Received 24 March 2012 balancing energy needs for highly and fully renewable European power systems for different levels and
Accepted 4 September 2012 mixes of wind and solar energy. By applying a dispatch strategy that minimizes the balancing energy
Available online 25 September 2012
needs for a given storage size, the interplay between storage and balancing is quantified, providing a
Keywords: hard upper limit on their synergy. An efficient but relatively small storage reduces balancing energy
Energy system design needs significantly due to its influence on intra-day mismatches. Furthermore, we show that combined
Large-scale integration with a low-efficiency hydrogen storage and a level of balancing equal to what is today provided by
Storage storage lakes, it is sufficient to meet the European electricity demand in a fully renewable power
system where the average power generation from combined wind and solar exceeds the demand by
only a few percent.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction
A fully renewable pan-European power system will depend on a
large share of non-dispatchable, weather dependent sources,
primarily wind and solar power (Czisch, 2005; Jacobson and
Delucchi, 2011). The optimal ratio between and necessary amount
of wind and solar power depends on storage and balancing resources
(Heide et al., 2010, 2011; Hedegaard and Meibom, 2012), transmis-
sion (Czisch et al., 2007; Kempton et al., 2010; Schaber et al., 2012a,b),
and the characteristics of the climate (Widén, 2011; Aboumahboub
et al., 2010) and load (Yao and Steemers, 2005). Based on
meteorological data, it was shown that even with unlimited trans-
mission within Europe, a scenario with only wind and solar power in
n
Corresponding author at: Center for Mathematical Sciences, Technische
Universität München, 85748 Garching, Germany.
E-mail addresses: mgr@imf.au.dk, mortengrud@gmail.com (M.G. Rasmussen).
combination with either only storage (Heide et al., 2010) or only
balancing (Heide et al., 2011) requires a very large amount of excess
generation in order to be technically feasible. Here, we study the
intermediate and more realistic scenarios where power generation
from the two weather-driven variable renewable energy (VRE)
sources is backed up by specific combinations of storage and
balancing. We identify a class of realistic and feasible scenarios for
building a fully or partially renewable pan-European power system.
The main point of our paper is to outline what is possible in a
wind and solar based European power system with storage and
balancing systems. The power capacities of the storage and
balancing facilities are not determined; this would require a more
complex modeling with explicit inclusion of power transmission
(Rodriguez et al., 2012). We focus on wind and solar power and
assume no bottlenecks in the power grid, employ an optimal
storage dispatch strategy and ignore storage charge and discharge
capacities and economic aspects. The incentives for doing so are
closely related and at least threefold.

0301-4215/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.enpol.2012.09.009
 M.G. Rasmussen et al. / Energy Policy 51 (2012) 642–651
By assuming few technical constraints and optimal operation
strategy, our results provide a hard upper limit on what can be
accomplished by better intercontinental power network integra-
tion, better technology or better market design. This means that
more realistic models have a solid frame of reference to be
measured up against. At the same time, our results provide
precise boundaries for policy makers. Furthermore, it allows for
a simpler analysis and a clearer presentation. The price is that a
more detailed model is necessary in order to give useful answers
to questions regarding e.g. charge, discharge,1 balancing or power
flow capacities, and hence any considerations regarding the
economic costs obviously become highly speculative within the
current modeling framework.
A somewhat related but in some respects also parallel argu-
ment is behind the decision of focusing on wind and solar power.
Three of the most scalable renewable energy sources are wind
and solar power and biomass. Of these, biomass stands out as
being dispatchable, meaning that it can be used for balancing the
intermittent (or non-dispatchable) power sources, such as wind
and solar power. As such, biomass is implicitly included in our
analysis. Besides wind and solar power, one of the most important
non-dispatchable power sources is run-of-river hydro power.
Although run-of-river contributes significantly to the present-
day power systems of Europe with some percent of the total
generation, the perspective for future growth is highly limited
(Lehner et al., 2005), leaving it ill-suited as a third component in
our parametric analysis. However, with available data, it would
have been straight-forward and an obvious choice to include it as
a fixed (but time dependent) contribution.
Contrary to most work done in this field (Czisch, 2005;
Jacobson and Delucchi, 2011; Martinot et al., 2007; Lund and
Mathiesen, 2009), the novel weather-based modeling approach,
used here and first introduced in Heide et al. (2010), is designed
to investigate a whole class of scenarios. This means that we do
not only provide results for a few end-points, or for a pathway to
any such. Rather, we provide a continuous map of results for any
combination and penetration level of wind and solar power
generation (Heide et al., 2010, 2011; Schaber et al., 2012b). We
do not attempt to show that a transition to a fully renewable
power system is economically viable, however, this has been
addressed by others (Czisch, 2005; ECF, 2010). An advantage of
disregarding the economical aspect is that our work is less
sensitive to technological advances and socio-economic changes
that could potentially shift the economic balance between e.g.
wind and solar power.
The paper is organized as follows: Section 2 describes data and
methodology. Section 3 briefly discusses the extreme scenarios of
no storage and sufficiently large storage, respectively. Section 4
deals with the interplay between storage and balancing. Section 5
treats the concrete example of having a 6-h storage. Section 6
discusses the impact of hydro balancing and large hydrogen
storages. Finally, the appendix discusses storage operation
strategies.
2. Data and methodology
2.1. Weather and load data
Historical weather data from the 8-year period 2000–2007
2
with a tempo-spatial resolution of 1 h  50  50 km was used to
derive wind and solar power generation time series for Europe
(see Heide et al., 2010). The data set comprise two 70,128-h long
1
See, however, the Appendix.
643
Fig. 1. Top panel shows excerpts of the L(t) (red), W(t) (blue) and S(t) (yellow)
time series, representing load and wind and solar power, respectively. Lower panel
shows mismatch time series DðtÞ for the same days (black), with g ¼ 1:00 and
aW ¼ 0:60. The blue areas indicate excess generation, while the red areas indicate
the residual load. The excerpt covers the time period 10–15 March 2000. (For
interpretation of the references to color in this figure caption, the reader is
referred to the web version of this article.)
time series for each of the approximately 2600 grid cells, covering
27 European countries including off-shore regions. In this study
all grid cells are aggregated using a capacity layout based on
political goals and attractiveness of the sites (see Heide et al.,
2010). We thus have two hourly power generation time series for
Europe, w(t) for wind and s(t) for solar power generation, where t
denotes any hour of the 8-year period.
The electrical load time series L(t) is based on data from the
transmission system operators (TSO) for the same 8 years. The
load data was de-trended to compensate for the approximately
2% yearly increase in demand and scaled to 2007 values, where
the total annual consumption amounted to 3240 TWh for Europe.
This corresponds to an average hourly load (av.h.l.) of 370 GWh.
The power generation time series w(t) and s(t) have been
normalized to the average electrical load, yielding two new time
series, W(t) and S(t). As everything is scaled to 2007 values, we
use the units av.h.l. (average hourly load) and av.y.l. (average
yearly load) throughout the paper. This means that
wðtÞ sðtÞ
WðtÞ ¼  av:h:l:, SðtÞ ¼  av:h:l:,
/wS /sS
and
/LS ¼ /WS ¼ /SS ¼ 1 av:h:l:,
where /  S denotes the average value of a time series. In these
units, the values of L(t) all lie in the range 0.62–1.47 av.h.l. while
W(t) and S(t) fall in the intervals 0.07–2.61 and 0.00–4.10 av.h.l.,
respectively. Fig. 1 shows an excerpt of the time series.
2.2. Generation–load mismatch
The generation–load mismatch time series DðtÞ is given by
DðtÞ ¼ g  ðaW  WðtÞ þð1aW Þ  SðtÞÞLðtÞ, ð1Þ
where aW denotes the wind power fraction of the average wind
and solar power generation and g represents how much combined
wind and solar power is generated on average, so that if e.g.
g ¼ 1:00, the average total generation of wind and solar power
equals the average load. Using the standard notation
( (
0 for x Z 0 x for x Z0
x ¼ , xþ ¼ ð2Þ
x for x o 0 0 for x o0,
this means that D ðtÞ denotes the residual load, and D þ ðtÞ
denotes possible excess generation. We will use the term
mix for any fixed choice of aW while g is referred to as the average
644 M.G. Rasmussen et al. / Energy Policy 51 (2012) 642–651
VRE2 generation factor. An excerpt of the mismatch time series DðtÞ
for an average VRE generation factor g ¼ 1:00 and a mix of
aW ¼ 0:60 together with the corresponding excerpt of the load
and wind and solar power time series can be seen in Fig. 1.
2.3. Balancing
As our focus is on wind and solar based, fully renewable power
systems, we consider any dispatchable, additional power source
to be balancing (excluding storage). Thus, when no storage is
present, the balancing time series is simply the residual load, or
minus the negative values of DðtÞ
BðtÞ ¼ D ðtÞ ðno storageÞ,
corresponding to the red areas in the lower panel of Fig. 1. In this
paper, we will mainly focus on one property of the balancing time
series, namely its average value (which we divide by the average
load to get a dimensionless parameter)
EB ¼ /BS=/LS,
EB measures the share of the energy that must be provided by
non-VRE resources to cover the demand. This means that one
would hope for EB þ g ¼ 1 for g r 1 and EB ¼0 for g 41, which can
be rewritten as
EB ¼ ð1gÞ þ ,
3
because otherwise, more energy is put into the system than is
demanded. This also means that, with a perfect storage, or with
perfect demand side management, (4) would hold (in fact, this
could be the definition of being perfect).
Unfortunately, (4) cannot be expected to hold. This means that,
for a given g, we could have EB 4 ð1gÞ þ , meaning that on
average, we need additional balancing. This leads us to define
the additional average balancing Eadd
B as
Eadd
B ¼ EB ð1gÞ þ :
One can think of Eadd
B as a measure of how well the VRE resources
can be integrated into the power system, or how much energy
one on average would need to move in time to get perfect
integration. We stress that Eadd is a property that has to do with
B t
an average—it is not a priori possible to assign a time series to the
additional balancing. As one always – given g – can find EB from
Eadd
B by adding ð1gÞ þ , and because Eadd B makes some of the
definitions related to storages easier, we will mainly use Eadd B storage filling level time series value Hðt max Þ, t max being the last
rather than EB, except when we deal with lossy storages, as the
concept of additional balancing in this case becomes more
complicated.
2.4. Storage
With our goal of outlining the borders of what is possible in a
European power system with storage and balancing, in particular
with respect to balancing energy EB, we will focus on a storage
dispatch strategy which, for a given storage size CS, performs
optimal with respect to EB. This means that given a storage size,
no storage dispatch strategy can result in a lower average
balancing energy EB than the one employed here. There are
several strategies that satisfy this condition, meaning that they
all result in the same EB.
Here, we will present a version which has the advantages that
it is simple and hence easily defined and is easily seen to satisfy
the condition, and it assumes no foresight capabilities, but has the
disadvantage that it is quite far from how present-day real-world
2
Variable renewable energy.
3
This would of course lead to curtailment.
operation works. Three other versions satisfying the minimum EB
condition are presented in the appendix.
The version presented here can be summed up as a ‘‘storage
first’’ strategy, in the sense that any deficits are first covered with
storage unless it runs empty, and any excess generation is stored
in the storage, unless the storage gets full. No limits are imposed
on the charge and discharge capacities. Conversion losses in and
out of the storage are modeled by storage efficiencies Zin and Zout . For
example, for hydrogen storage, the efficiencies are approximately
0.60 in both directions (Beaudin et al., 2010; Kruse et al., 2002),
leading to a round-trip efficiency of 0.36.
A storage is thus characterized by three parameters: Zin , Zout
and CS. The storage filling level time series H(t) with a constrained
ð3Þ
storage size CS is given by
8 ~ ðtÞ 4C S ,
< CS
> for Hðt1Þ þ D
HðtÞ ¼ 0 ~ ðtÞ o0,
for Hðt1Þ þ D ð5Þ
>
:
~ ðtÞ
Hðt1Þ þ D otherwise,
~ is given by the equation
where D
D~ ðtÞ ¼ Zin D þ ðtÞZ1
out D ðtÞ, ð6Þ
and Hðt min 1Þ ¼ H0 is an initial value to be determined, t min being
the first hour of the time series. The storage works in the
ð4Þ
following way: any excess power at any time is fed into the
storage with an efficiency of Zin , unless the storage size is
exceeded, in which case the storage is full. Any deficits in VRE
power generation as compared to the load are covered by the
storage with an efficiency of Zout , except if the storage runs empty,
in which case the storage only provides partial coverage of the
deficit. The dispatch strategy is illustrated in Fig. 2.
To ensure storage neutrality, i.e. that the storage provides
only as much energy as is stored, we determine H0 in the
following way: first, it is examined whether the generation
can match the demand. This is the case if D ~ sums up to a non-
P ~
negative value, t D ðtÞ Z 0 (equality is called equilibrium) or
equivalently
X X
Zin Zout D þ ðtÞ Z D ðtÞ: ð7Þ
t
If this is the case, we set the temporary variable H00 ¼ C S , if not,
H00 ¼ 0. H00 is then used as an initial guess for H0, and the final
hour of the time series, is found. We then choose H0 ¼ Hðt max Þ.
With this choice, the storage becomes neutral, which means that
Fig. 2. Top panel shows excerpt of a storage filling level time series resulting from
applying the simple dispatch strategy. Lower panel shows excerpt of mismatch
(outer black line) for the time series leading to the storage filling level time series
in the top panel, as well as charge (dark green), discharge (light green), excess
(blue) and balancing power (red), respectively. The excerpt of the time series
shown is the time period 10–15 March 2000. (For interpretation of the references
to color in this figure caption, the reader is referred to the web version of this
article.)
 M.G. Rasmussen et al. / Energy Policy 51 (2012) 642–651 645

A case of more than one storage, e.g. a small high-efficient

short-term and a larger low-efficient seasonal storage, can be
modeled by feeding the reduced mismatch time series resulting
from the application of the smaller storage to the model of the
larger storage (see Sections 5 and 6).

3. The no storage and no additional average balancing

scenarios

3.1. The no-storage scenario

Heide et al. (2011) have shown that in a scenario without storage

Fig. 3. Top panel shows as an example the storage filling level time series
and with g Z1 the balancing needs are significant, even when the
resulting from g ¼ 0:9 and aW ¼ 0:6 with a lossless, unconstrained storage. As
g o 1, the time series tends to be close 0. C eff corresponds to the highest peak. average VRE generation exceeds the load, and when aggregated over
S
Lower panel similarly shows the storage filling level time series resulting from all of Europe. With the improved model presented here, we extend
g ¼ 1:1 and a ¼ 0:6. As g 4 1, the time series tends to be close to C eff
S , which in turn this scenario to the regime g o 1.
is determined by the largest dip.
As revealed in Fig. 4 the integration of wind and solar power
works well up to around g ¼ 0:5. Almost no additional average
balancing is required up to this penetration. Above this limit large
the storage does not contribute to the balancing, nor does it
amounts of additional average balancing are required. Some of
consume extra energy over the 8-year time span considered.
the generated power is potentially lost as excess generation, if it is
In case the whole storage range is not used, the storage filling
not stored or if a new use for highly intermittent power is not
level time series is shifted down so that the minimal value of the
found. The additional average balancing fraction Eadd B strongly
storage filling level becomes 0. For a lossless storage ðZin ¼
peaks at g ¼ 1. The size of the peak strongly depends on the
Zout ¼ 1Þ, this happens if the storage is large enough to avoid mix aW . For aW ¼ 0:8 it becomes a minimum. Also for g a1 the
additional average balancing. The effectively used storage size C eff
S mix aW ¼ 0:8 remains close to optimal. The optimal aW minimizes
is then given by maxt HðtÞ. By choosing CS sufficiently large,
balancing energy when no storage is present, c.f. Heide et al.
making the storage size in practice unconstrained, the algorithm
(2011), and turns out to be g dependent. The numerical value of
can in this way be used to determine the smallest storage size
the optimal mix changes from 0.82 at g ¼ 0:5 via 0.80 at g ¼ 1:00
with the property that enlarging the storage size has no effect on
to 0.88 at g ¼ 2:00.
what can be stored or dispatched. Assuming lossless storage with
Zin ¼ Zout ¼ 1:00, the model with unconstrained storage size
amounts to storing all excess for g r1, and exactly what is 3.2. The no additional average balancing scenario
required to cover deficits for g Z1. Fig. 3 illustrates the storage
filling level time series for unconstrained storage sizes. As already explored for g Z1 in Heide et al. (2010, 2011), the
alternative scenario where additional average balancing is com-
2.5. Reduced mismatch time series pletely avoided by storing enough excess generation to cover
additional average balancing needs, leads to very large storage
So far, we have defined the generation and load time series, sizes. Using the extended model, we can also cover g o 1.
P ~
the mismatch time series, the balancing time series and (addi- At equilibrium ð t D ðtÞ ¼ 0Þ, which for lossless storages is reached
tional) average balancing energy as well as the storage filling level at g ¼ 1:00, a pronounced cusp singularity of required storage size
time series. To model the interaction of the mismatch with the appears. See Fig. 5.
storage, we define a reduced mismatch time series, from which we
can determine how storage affects balancing needs.
The changes in the filling level of the storage can be described
by the time series F(t) given by
FðtÞ ¼ HðtÞHðt1Þ, t Zt min : ð8Þ
Correspondingly, the power flow in and out of the storage is
F~ ðtÞ ¼ Z1
in F þ ðtÞZout F  ðtÞ, ð9Þ
where positive values indicate that power flows in and negative
values indicate that power flows out of the storage. We can now
write the mismatch after storage transactions as
Dr ðtÞ ¼ DðtÞF~ ðtÞ, ð10Þ
where r stands for ‘‘reduced.’’ The flows as well as the reduced
mismatch is illustrated in Fig. 2 for a lossless storage. The reduced
mismatch is now used to find the balancing time series for
scenarios with combined storage and balancing, simply by
BðtÞ ¼ Dr ðtÞ: ð11Þ
Fig. 4. Additional average balancing EaddB vs. average VRE generation factor g
This definition replaces the previous definition (3) of the balan- without storage for aW ¼ 0:60 (blue), aW ¼ 0:80 (red) and the g dependent optimal
cing time series. It coincides with (3) when CS ¼0. The definitions aW (gray). (For interpretation of the references to color in this figure caption, the
of EB and Eadd
B are unchanged. reader is referred to the web version of this article.)
646 M.G. Rasmussen et al. / Energy Policy 51 (2012) 642–651

Fig. 6. Additional average balancing fraction Eadd B as a function of small storage

Fig. 5. Storage singularity: required storage size C eff
S without additional average size CS (shown on a linear scale). The darker curves are for aW ¼ 0:60 and g ¼ 0:75
balancing vs. average VRE power generation factor g for aW ¼ 0:60 (blue),
(red), 1.00 (green), 1.25 (blue). The light curves are obtained with an optimal mix
aW ¼ 0:80 and the g-dependent optimal aW (gray). (For interpretation of the which minimizes the additional average balancing fraction (see Fig. 7). The inset
references to color in this figure caption, the reader is referred to the web version
shows Eadd
B as a function of large storage sizes CS (now shown on a log-scale). Note,
of this article.)
that the endpoints of the curves reflect the findings of Figs. 4 and 5. (For
interpretation of the references to color in this figure caption, the reader is
The choices of mix in Fig. 5 reflect the findings of Heide et al. referred to the web version of this article.)
(2010, 2011), with the mix aW ¼ 0:60 minimizing storage at
g ¼ 1:00 and the mix aW ¼ 0:80 minimizing balancing at
g ¼ 1:00. Here, the optimal aW is the g dependent choice that
minimizes the storage size without introducing additional aver-
age balancing. This optimal mix varies between 0.55 and 0.78 and
is different from the optimal mix discussed in the previous
subsection.
As seen in the figure, also for g o1 the storage size resulting
from the mix aW ¼ 0:60 is very close to the result for the optimal
mix, and significantly better than that of the mix aW ¼ 0:80. When
g gets larger than 1, the aW ¼ 0:60 and aW ¼ 0:80 cases quickly get
very close to each other, reflecting the fact that, as excess
generation increases, it is easier to keep the storage filled up
and deficits become more rare, regardless of the mix. The case of
optimal mix, however, performs slightly better in the g range
1.00–1.50.

4. Combined usage of lossless storage and balancing

In this section, we investigate the advantages of combining

storage and balancing. The scenarios described in Section 3 are
extreme cases of a more general setup where storage and
balancing are used in combination. In between the extremal
balancing and storage scenarios, there is an unlimited number
of possible combined balancing and storage strategies. In this
paper, we will restrict ourselves to storage first strategies, with an
imposed limit on the storage size, and balancing handling any
remaining mismatch as described in Section 2. This class of
strategies can be parametrized by the given storage size. They
include the extremal balancing and storage scenarios as special
cases by setting the imposed storage size limit to 0 or sufficiently
large, respectively. A key property of the storage first strategies is
that they minimize the additional average balancing fraction for a
given storage size. See Appendix A for a brief discussion of these
and other strategies.
Fig. 6 shows the relation between storage size and additional
average balancing fraction. The figure shows three groups of
curves. The red ðg ¼ 0:75Þ, green ðg ¼ 1:00Þ and blue ðg ¼ 1:25Þ
are for the mix aW ¼ 0:60. The respective light-colored curves are
obtained with an optimal mix, which minimizes Eadd B for a given
Fig. 7. Optimal mix as a function of storage size for g ¼ 0:75 (red), 1.00 (green),
1.25 (blue). The respective dotted curves represent the upper and lower limits of
aW , which are obtained by allowing the additional average balancing fraction to be
one percentage point larger than the minimized Eadd B . For comparison the choice
aW ¼ 0:60 is indicated by a gray line. (For interpretation of the references to color
in this figure caption, the reader is referred to the web version of this article.)
storage size CS and which differs slightly from aW ¼ 0:60, espe-
cially for very small storage sizes CS; see also Fig. 7. As can be
seen, all six curves Eadd
B ðC S Þ are strongly convex, meaning that the
reduction in balancing per unit of extra storage is much greater at
the outset than for subsequent extensions of the storage size.
There is a huge decrease in additional average balancing fraction
for rather small storage sizes up to 6 av.h.l. (a quarter of a day).
Beyond 6 av.h.l. of storage size the decrease of additional average
balancing fraction is slow.
The different behavior of the additional average balancing
fraction in the two regimes with storage size below and above
6 av.h.l. is easily explained. Caused by the solar power generation
and the load, mismatch (1) shows a pronounced diurnal pattern.
For not too large mixes aW the mismatch is mostly positive during
 M.G. Rasmussen et al. / Energy Policy 51 (2012) 642–651 647

day-light hours and negative during the night. A 6 av.h.l. storage is

able to smooth this intraday cycle. It stores the positive mismatches
occurring during the day and serves the negative mismatches
occurring during the night; see also Heide et al. (2011). This explains
the fast decrease of the excess balancing fraction with the introduc-
tion of a small storage size. How good the 6-h storage is at smoothing
the intraday cycle clearly depends on how much solar power is
assumed to be in the system. If the solar share is very large, the
6 h become inadequate and should be increased to about 12 h in
the extreme case aW ¼ 0. Besides the diurnal time scale, the mis-
match fluctuations are also influenced by the larger synoptic and
seasonal time scales. The latter two cause the slow decrease of the
additional average balancing fraction for storage sizes beyond 6 av.h.l.
Note that the optimal mix between wind and solar power
generation is also different for the two storage regimes (c.f. Fig. 7).
Sufficiently below the 6 av.h.l. storage size, the optimal mix is
aW C 0:8, and the balancing energy increases relatively rapidly, as
one deviates from the optimum value, as indicated by the dotted
lines in Fig. 7. This is a consequence mostly of the fact that the strong
diurnal patterns of solar power makes it difficult to integrate large
quantities of it into the power system without storage. At 6 av.h.l. the
optimal mix has decreased to aW ¼ 0:520:6 and the range of mixes
that result in approximately the same balancing has increased
significantly. For larger storage sizes the mix 0.6 results in a
balancing energy close to the optimal mix for all g’s. In this regime
the optimal mix is no longer sensitive to fluctuations on the diurnal
time scale. It is dominated by the fluctuations occurring on the
seasonal time scale. We note that the optimal mix aW ¼ 0:60 differs
from what is indicated in Schaber et al. (2012b) where a mix of 0.80
is suggested to be optimal in the presence of storage and a large VRE
penetration.4 As the aW ¼ 0:60 curves are good approximators of the
optimal mix curves for storages larger than 6 av.h.l., we will focus on
this choice in what follows.

5. A lossless 6-h storage with unconstrained balancing

and seasonal storage
As described above, the introduction of a 6-h storage has a
significant impact on the required balancing energy as compared
to no storage. Contrary to the enormous storage sizes needed in
the full storage scenario (c.f. Fig. 5), a 6-h storage also appears to
be technically feasible. Storages of that order may be realized by a
combination of many different storage and time-shift technolo-
gies such as pumped hydro, compressed air, superconduc-
ting magnetic energy storages, different battery technologies,
flywheels, capacitors (Beaudin et al., 2010) and demand side
management and smart grid (Strbac, 2008).
Fig. 8a plots the additional average balancing fraction vs.
average VRE generation factor for the mix aW ¼ 0:60. The black
curve is with a lossless 6-h storage while the gray curve
is without storage. The additional average balancing fraction is
more than halved at g ¼ 1:00, where the difference is smallest in
relative terms. For other values of g the relative reduction in
additional average balancing is larger. Whereas without storage
significant amounts of additional average balancing appear
beyond g around 50%, this threshold is pushed to 75% when
introducing the 6-h storage. As for the other end of the range,
where 3:9% of additional average balancing is needed at g ¼ 2:00,
this is reduced to just 0:1% with a 6-h storage.
4
This due to the fact that the measure used in Schaber et al. (2012b), which
P
can be defined as D ¼ t 9DðtÞ9, has no temporal memory. This means that,
in principle, the storage could be full all summer and empty all winter, resulting
in 1 yearly storage cycle, and still result in a small D.
Fig. 8. (a) Additional average balancing fraction Eadd and (b) storage size C eff
B S for a
hydrogen storage vs. average VRE power generation factor g with (black) and
without (gray) first using a 6-h lossless storage for aW ¼ 0:60.
Another option is to have two layers of storage: the high-
efficient 6-h storage and a low-efficient bulk seasonal storage.
Additional storages with sizes far beyond 6 av.h.l. require high
energy densities. An option could be hydrogen storage
(Hedegaard and Meibom, 2012; Beaudin et al., 2010; Kruse
et al., 2002). We first compute the reduced mismatch time series
after applying the 6-h storage to the generation–load mismatch
time series, and the lower efficiency storage is then applied to this
reduced mismatch time series. Since all that can be covered by
the 6-h storage is assumed to be lossless, we increase the overall
energy efficiency by applying this first. This can be directly
observed in Fig. 8b, where the size for the less efficient hydrogen
storage is shown as a function of g with and without the 6-h
lossless storage for the mix aW ¼ 0:60. Without the 6-h storage,
the size of the bulk hydrogen storage reaches equilibrium
P ~
( tD ðtÞ ¼ 0, see Section 2.4) at the peak g ¼ 1:23. The 6-h storage
lowers this value down to g ¼ 1:10 (here the equilibrium is at
P ~r
t D ðtÞ ¼ 0).
6. A lossless 6-h storage with constrained hydro balancing
and hydrogen storage
A possible way of providing a renewable form of balancing
that can be dispatched on demand is to use the hydro power
648 M.G. Rasmussen et al. / Energy Policy 51 (2012) 642–651

storage lakes of northern Scandinavia, the Alps and elsewhere in

Europe. In France, Italy, Spain and Switzerland alone, the current
generation from storage lake facilities amounts to about
70 TWh=year, and the total hydro power generation of Norway
and Sweden amounts to 190 TWh=year, most of which is gener-
ated from storage lake facilities (Lehner et al., 2005). Here, we
assume that at least 150 TWh=year of the total European hydro
power generation can be dispatched on demand. This amounts to
about 5% of the average load, and to the best of our knowledge, it
represents a conservative estimate. As an alternative to hydro
power, other sources of dispatchable renewable balancing such as
biomass fired power plants could also be assumed without
changing the conclusions of the analysis.
Bulk seasonal storage in a fully renewable pan-European power
system could be realized as e.g. underground hydrogen storage.
Of the different types of underground storage types presently used
for natural gas, only solution mined salt caverns are directly usable
for hydrogen storage (Stone et al., 2009). In Europe most of the
suitable geological formations are located in northern Germany, and
the combined working gas volume of all existing European facilities
allows for storage of 32.5 TWh in the form of hydrogen (Gilhaus,
2007). In the following we assume that a total storage size of
25 TWh are made available for hydrogen storage and we use
conversion efficiencies of Zin ¼ Zout ¼ 0:60.
With a total balancing energy of 150 TWh/year and a 25 TWh
hydrogen storage, as described above, we model the optimal
combined operation with and without an additional lossless 6-h
storage (2.2 TWh). As outlined at the end of Section 2, first the
high-efficient 6-h storage, if included, is applied to the mismatch
time series, next the reduced mismatch time series is confronted
with the constrained hydrogen storage, and finally, the still
existing (doubly reduced) mismatch determines the balancing
needs. According to Fig. 8b, for the mix aW ¼ 0:60 the storage size
of 25 TWh (corresponding to 0.008 av.y.l. in 2007-units) only
represents a constraint for average VRE power generation factors
g between 0.97 and 1.80 without a 6-h storage and between 0.92
and 1.55 with a 6-h storage.
Now the question is which combinations of g and aW satisfy
the 150 TWh=year constraint of hydro balancing. The answer to
Fig. 9. Contour plots indicating which g–aW combinations lead to a balancing
this question is given in Fig. 9 for the six scenarios: no, con- need equal to 150 TWh=year for the three scenarios (blue) no, (green) constrained
strained ð25 TWhÞ and unconstrained hydrogen storage, com- (with 25 TWh) and (red) unconstrained (with effectively unlimited size) hydrogen
bined with and without a lossless 6-h storage. The figure shows storage. The top and bottom panels are, respectively, without and with a 6-h
the penetration levels and mixes that qualify as fully renewable. lossless storage applied first. On both panels, the solid lines indicate where
150 TWh=year balancing suffice, and the dash-dotted and dashed lines indicate
We note that a scenario cannot qualify if the combined renewable
the corresponding lines for 150 TWh=year 7 10%, respectively. (For interpretation
power generation from wind, solar PV and hydro balancing does of the references to color in this figure caption, the reader is referred to the web
not equal or exceed the load on average. In the case studied here, version of this article.)
hydro balancing corresponds to 0.046 av.h.l. so the minimum g
is 0.954 av.h.l. However, because some energy is lost when it
is stored as hydrogen, and some wind and solar PV is curtailed, impact of a lossless 6-h storage is comparable to that of a 25 TWh
the actual minimum value of g is higher. hydrogen storage, albeit at a very different mix. When a 25 TWh
Fig. 9a shows the results of calculations without the lossless 6-h hydrogen storage is added too, the optimal mix becomes
storage. When none of the storages are present, the optimal mix is aW ¼ 0:56, and we find that g ¼ 1:03 suffice. In this case, increas-
aW ¼ 0:85 and the minimum value of g is 1.50. The resulting ing the hydrogen storage size beyond 25 TWh only has a marginal
average total renewable generation is about 1.55 av.h.l. since hydro effect.
balancing must be added to generation from wind and solar PV. In In the scenario where a 6-h storage is added to the constrained
this scenario, the entire 0.55 av.h.l. is lost to curtailment. Adding the hydrogen storage, the total storage size is increased from 25 to
constrained 25 TWh hydrogen storage alone reduces the minimum 27.2 TWh. However, only a small part of the reduction in g from
value of g to 1.15, now with aW ¼ 0:74. An unconstrained hydrogen 1.15 to 1.03 can be attributed to this increase. Instead, the more
storage could reduce the value of g further to 1.09 at an aW of 0.81. important reasons for the decrease are: (i) a higher total energy
In the two scenarios with either constrained or unlimited hydrogen efficiency, where the lossless 6-h storage allows efficient smooth-
storage, the overproduction is lost to a combination of storage ing of the intra-day variations in the mismatch and (ii) it enables
conversion losses and curtailment. more efficient use of solar PV and, thus, a seasonal optimal mix
Fig. 9b shows how the three scenarios, described above, between wind and solar PV can be better utilized. As a result, the
change when a lossless 6-h storage (2.2 TWh) is added to the scenario which includes both the constrained hydrogen storage
system. Now, a scenario without hydrogen storage can be made and the lossless 6-h storage results in a combined average
fully renewable at a g of 1.12 and an aW ¼ 0:56 mix. Thus, the generation that corresponds to only ð3þ 5Þ% ¼ 8% more than
 M.G. Rasmussen et al. / Energy Policy 51 (2012) 642–651
the average demand. Thus, this scenario has a high over-all
energy efficiency and a low amount of overinstallation is needed.
As such it is an example of a fully renewable European electricity
supply based on wind, solar PV and a conservative estimate of
hydro power generation alone.
7. Discussion and conclusion
We have investigated the ability of the power system to
integrate any level of wind and solar power in a fully intercon-
nected European power system. In our analysis, we have focused
on the ability of both short-term and seasonal storage systems to
increase the integrability. Our study addresses how the introduc-
tion of a certain amount of storage influences (1) the balancing
energy needs and (2) the optimal generation mix. The analysis is
based on high-resolution hourly weather data and electricity load
for 27 European countries over the 8-year period 2000–2007.
Without additional balancing, a very pronounced cusp singu-
larity of the needed storage size appears, peaking where the
system is fully renewable (c.f. Figs. 5 and 8b). This is due to the
fact that the storage filling level time series changes from being
almost empty most of the time to being almost full most of the time,
resembling a phase-transition. At its peak, the storage size amounts to
at least 320 TWh or 10% of the annual load. For hydrogen storage and
with suboptimal mix, the figure is higher. The technical implementa-
tion and physical location of such a storage is severely limited and
may not be feasible (Heide et al., 2010). With additional balancing,
the need for such a large storage can, however, be avoided.
For storages of limited size, we have identified that a storage
capable of storing 6 h of average consumption – a quarter of an
average day – is enough to remove the largest impact of the
intraday mismatch, and thus significantly reduce the balancing
energy needs (c.f. Fig. 8). This implies that the optimal mix is
shifted towards the seasonal optimal mix. A high-efficiency 6-h
storage more than halves the balancing energy needs at g ¼ 1:00
for a mix of aW ¼ 0:60. The impact is even greater for both higher
and lower penetration levels, allowing for full integration up to an
average combined wind and solar power generation of about 75%
(c.f. Fig. 8a). Without the 6-h storage, this full integration can only
be achieved up to about 50% for a fully connected Europe.
As indicated in Fig. 9, the impact of a 6-h lossless storage (at
optimal mix) is under certain circumstances almost as good as an
unlimited hydrogen storage (at optimal mix). As opposed to the
320 TWh storage described above, a 6-h storage, which corre-
sponds to 2.2 TWh, can be realized by combining many different
technological solutions including physical storages and time-shift
technologies. Our results indicate that political support for the
development of such technologies will be very beneficial for the
system when the VRE penetration reaches 50% and will continue
to have a great impact even in the fully renewable regime.
A seasonal storage can most likely only be realized with a low
round-trip efficiency. A low-efficiency storage with a round-trip
efficiency of 0.36 (0.60 each direction) and a storage size of
25 TWh (capable of providing up to 0:60  25 TWh ¼ 15 TWh of
stored energy), is technically realizable e.g. as hydrogen stored in
solution mined salt caverns in northern Germany. In this scenario,
the high-efficiency 6-h storage serves the purpose of reducing the
conversion losses, making the 150 TWh=year of balancing suffi-
cient for a g of just 1.03 at an aW ¼ 0:56 mix (again, c.f. Fig. 9).
Without the 6-h storage, the corresponding number would be
g ¼ 1:15 ðaW ¼ 0:80 mix). Two fully renewable scenarios can be
realized with a lossless 6-h storage: one with g ¼ 1:03 and a
25 TWh hydrogen storage and one with g ¼ 1:12 and no addi-
tional storage. Whether there should be a hydrogen storage
depends on the price of the 25 TWh compared to the price of
649
going from g ¼ 1:03 to g ¼ 1:12, be it measured in environmental
impact, economical terms or otherwise. If a highly efficient short-
term storage comparable to our 6-h storage does not become
feasible, the impact of a hydrogen storage is much greater, as it
reduces the needed g from 1.52 to 1.15 or 1.09, depending on
whether the storage is limited to 25 TWh or not.
In conclusion, we find a significant synergy between storage
and balancing. In particular, the effect of a highly efficient short-
term storage dramatically reduces the balancing energy needs
and allows for efficient use of a mix close to the seasonal optimal
mix of wind and solar power. However, the additional gain by
increasing the storage size of such a storage further is limited.
A seasonal storage will most likely have a low efficiency, but in
combination with a highly efficient short-term storage, the
efficiency of the combined storage system is increased. This
reduces the needed amount of overproduction in a fully renew-
able scenario. With a balancing of only 150 TWh=year, e.g. coming
from biomass or hydro power, a highly efficient 6-h storage and a
25 TWh hydrogen storage employed, we find that a fully renew-
able scenario can be realized with an average wind and solar
power production of only 3% more than the average load.
Increasing the hydrogen storage size further does not lower the
needed amount of overproduction substantially, and increasing
the figure to 12%, a hydrogen storage can be completely avoided.
At present, the intracontinental power grid is becoming a
bottleneck for the integration of non-dispatchable renewables
such as wind and solar power. We find that even with a perfect
grid, a combined penetration of wind and solar power of about
50% will lead to the need for an energy storage in order to avoid
large losses. Investment in an efficient (virtual or physical)
storage able to store the average demand for 6 h is very important
as it bears large benefits. However, we find that building a storage
large enough to handle all surplus generation at large penetra-
tions is unfeasible. A moderately large storage provides almost
the same benefits, in particular in combination with the highly
efficient small storage.
Acknowledgments
We would like to thank Rolando A. Rodriguez for proofreading
several earlier versions of this manuscript and the two anon-
ymous referees for comments that markedly improved the quality
of this final paper. The first author was supported by the Carlsberg
Foundation during all stages of the work. The second author is
supported by DONG Energy.
Appendix A. Strategies
In this appendix, we discuss the optimality of the ‘‘storage first’’
strategy with regard to balancing minimization for a given storage
size and give a few examples of other strategies with other
optimality properties. The fact that the ‘‘storage first’’ strategy is
indeed optimal with respect to minimizing balancing energy follows
from a simple induction argument, which is left to the interested
reader as an easy exercise for the mathematically trained.
We stress that this strategy is not the only optimal strategy in this
respect. Recall that the presented storage first strategy acts on an
hour-by-hour basis using the storage to cover deficits in the
generation–load mismatch time series if the storage is not empty,
and balancing in case the storage is or runs empty, and in case of a
positive mismatch, the excess generation is put into the storage if the
storage is not already full. This way of acting makes no use of
forecasts. Assuming perfect forecast, however, the same action could
be performed on larger time intervals, in our case same-sign
650 M.G. Rasmussen et al. / Energy Policy 51 (2012) 642–651
intervals, i.e. we split the time series in the largest chunks of intervals
where the mismatch time series does not change sign. Having done
this, one can now check to see how much of the integral of a positive
mismatch interval can fit into the storage and how much of the
integral of a negative mismatch interval can be covered by what is in
the storage. Depending on what one wants to reduce (excess,
balancing, storage charge/discharge power quantiles), one can then
either cover the negative mismatches using the storage to cover the
most negative part of the mismatch (trough filling) or using the
storage to cover with a more or less constant storage power output
(top filling), and vice versa for the positive mismatches (peak shaving
and bottom shaving, respectively). In either case, the energy sum of
what is used or absorbed by the storage in any given same-sign
interval remains the same as for the storage first strategy, leaving the
minimization with respect to balancing energy unaffected. The
original version (Strategy 1—‘‘storage first’’) and two combinations
(peak shaving–trough filling (Strategy 2) and bottom shaving–top
filling (Strategy 3), respectively) are illustrated in Fig. A1.
Fig. A1. Storage filling level (top panels) and mismatch, storage flow, excess and
balancing power (bottom panels) for Strategies 1–3, respectively. The positive
mismatch is divided into a dark green and a blue part: dark green is stored, while
blue is excess power not absorbed by storage. Likewise, the negative mismatch is
divided into a light green and a red part: light green is covered by storage while
the red represents coverage by balancing. The excerpt of the time series shown is
the time period 10–15 March 2000. (For interpretation of the references to color in
this figure caption, the reader is referred to the web version of this article.)
Fig. A2. Storage filling level (top panels) and mismatch, storage flow, excess and
balancing power (bottom panels) for Strategies 4 and 5, respectively. See Fig. A1
for explanation of the color coding.
Applying these three strategies to the same mismatch time
series clearly result in the same balancing energies, but the
quantiles of the storage charge and discharge powers and balan-
cing power differ significantly—in particular for large g’s. Even
minor changes in the storage–balancing strategies can have
significant impact on the storage charge and discharge powers
and the balancing power quantiles, and hence the needed
installed power capacities.
Another strategy, which we call Strategy 4, that leads to the
same balancing energy but a reduced charge and discharge
capacity is found by the following algorithm: First, Strategy 1 is
applied to determine how low the balancing energy can get for
the given CS. Then the charge capacity is reduced until it results in
a larger balancing energy as compared to the unconstrained
charging capacity. Now the charge capacity is kept fixed at the
least capacity that does not affect the balancing energy, while the
discharge capacity is reduced until it results in larger balancing
energy.
A suboptimal strategy along the same lines, Strategy 5, can be
defined as Strategy 4, but with the difference that a slightly higher
balancing energy, say 10% larger, is allowed. Strategies 4 and 5 are
illustrated in Fig. A2. In this concrete example, where the
balancing energy is allowed to be 10% larger (EB ¼ 0.1 instead of
EB ¼0.09) for Strategy 5, g ¼ 1:0 and aW ¼ 0:6, the storage charge
and discharge capacities are both approximately halved as com-
pared to Strategy 4, and compared to the other strategies, the
difference is even larger.
We will not pursue this issue any further in this paper, as a
more complex model including transmission and storage sites
seems to be needed in order to give useful results.
References
Aboumahboub, T., Schaber, K., Tzscheutschler, P., Hamacher, T., 2010. Optimal
configuration of a renewable-based electricity supply sector. WSEAS Transac-
tions on Power Systems 5, 120–129.
 M.G. Rasmussen et al. / Energy Policy 51 (2012) 642–651
Beaudin, M., Zareipour, H., Schellenberglabe, A., Rosehart, W., 2010. Energy storage
for mitigating the variability of renewable electricity sources: an updated
review. Energy for Sustainable Development 14, 302–314.
Czisch, G., 2005. Szenarien zur zukü nftigen Stromversorgung, kostenoptimierte
Variationen zur Versorgung Europas und seiner Nachbarn mit Strom aus
erneuerbaren Energien. Ph.D. Thesis. Universität Kassel.
Czisch, G., Giebel, G., 2007. Realisable scenarios for a future electricity supply
based 100% on renewable energies. In: Energy Solutions for Sustainable
Development Proceedings Risø International Energy Conference, Risø National
Library, Technical University of Denmark, pp. 186–195.
ECF, 2010. Roadmap 2050: A Practical Guide to a Prosperous, Low-carbon Europe.
Report. European Climate Foundation.
Gilhaus, A., 2007. Natural Gas Storage in Salt Caverns—Present Status, Develop-
ments and Future Trends in Europe. Technical Conference Paper. Solution
Mining Research Institute.
Hedegaard, K., Meibom, P., 2012. Wind power impacts and electricity storage—a
time scale perspective. Renewable Energy 37, 318–324.
Heide, D., von Bremen, L., Greiner, M., Hoffmann, C., Speckmann, M., Bofinger, S.,
2010. Seasonal optimal mix of wind and solar power in a future, highly
renewable Europe. Renewable Energy 35, 2483–2489.
Heide, D., Greiner, M., von Bremen, L., Hoffmann, C., 2011. Reduced storage and
balancing needs in a fully renewable European power system with excess
wind and solar power generation. Renewable Energy 36, 2515–2523.
Jacobson, M.Z., Delucchi, M.A., 2011. Providing all global energy with wind, water,
and solar power, Part I: technologies, energy resources, quantities and areas of
infrastructure, and materials. Energy Policy 39, 1154–1169.
Kempton, W., Pimenta, F.M., Veron, D.E., Colle, B.A., 2010. Electric power from
offshore wind via synoptic scale interconnection. Proceedings of the National
Academy of the Sciences of the United States of America 107, 7240–7245.
651
Kruse, B., Grinna, S., Buch, C., 2002. Hydrogen—Status og Muligheter. Technical
Report 6. Bellona (in English).
Lehner, B., Czisch, G., Vassolo, S., 2005. The impact of global change on the
hydropower potential of Europe: a model based analysis. Energy Policy 33,
839–855.
Lund, H., Mathiesen, B.V., 2009. Energy system analysis of 100% renewable energy
systems—the case of Denmark in years 2030 and 2050. Energy 34, 524–531.
Martinot, E., Dienst, C., Weiliang, L., Qimin, C., 2007. Renewable energy futures:
targets, scenarios, and pathways. Annual Review of Environment and
Resources 32, 205–239.
Rodriguez, R., Andresen, G.B., Becker, S., Greiner, M., 2012. Transmission needs in a
fully renewable pan-European electricity system. In: Uyar, T.S., Sağlam, M.,
Sulukan, E. (Eds.), 2nd International 100% Renewable Energy Conference and
Exhibition (IRENEC 2012) Proceedings, Maltepe, Istanbul, Turkey, pp. 320–324.
Schaber, K., Steinke, F., Hamacher, T., 2012a. Transmission grid extensions for the
integration of variable renewable energies in Europe: who benefits where?.
Energy Policy 43, 123–135.
Schaber, K., Steinke, F., Mühlich, P., Hamacher, T., 2012b. Parametric study of
variable renewable energy integration in Europe: advantages and costs of
transmission grid extensions. Energy Policy 42, 498–508.
Stone, H.B.J., Veldhuis, I., Richardson, R.N., 2009. Underground hydrogen storage in
the UK. The Geological Society, London, Special Publications 313, 217–226.
Strbac, G., 2008. Demand side management: benefits and challenges. Energy Policy
36.
Widén, J., 2011. Correlations between large-scale solar and wind power in a future
scenario for Sweden. IEEE Transactions on Sustainable Energy 2, 177–184.
Yao, R., Steemers, K., 2005. A method of formulating energy load profile for
domestic buildings in the UK. Energy and Buildings 37, 663–671.